5-fold cross-validation is a model evaluation technique in which the available dataset is partitioned into five equal, mutually exclusive subsets (folds), with the model trained on four folds and tested on the remaining one in five successive iterations, ensuring every sample is used for both training and testing exactly once. In WiFi CSI sensing research, this method is critical for producing reliable and generalizable performance estimates of classifiers — such as those used in human activity recognition or occupancy estimation — particularly when labeled CSI datasets are limited in size due to the practical difficulties of large-scale data collection in real environments. A common variant is stratified 5-fold cross-validation, in which class proportions are preserved across folds to prevent evaluation bias in imbalanced datasets, such as those where certain activities or occupancy states are underrepresented.
Source Papers
- Guiding Wi-Fi Sensor Placement for Enhanced CSI-Based Sensing in Stationary Crowd Counting ↗ — Guiding Wi-Fi Sensor Placement for Enhanced CSI-Based Sensin
- Human Activity Recognition via Wi-Fi and Inertial Sensors With Machine Learning ↗ — Human Activity Recognition via Wi-Fi and Inertial Sensors Wi
- Implementing Wi-Fi CSI-based room-level occupancy Estimation: an experimental study in multi-zone residential environments ↗ — Implementing Wi-Fi CSI-based room-level occupancy Estimation